It is currently Thu Mar 28, 2024 11:29 am

All times are UTC - 8 hours [ DST ]




Post new topic Reply to topic  [ 67 posts ]  Go to page Previous  1, 2, 3, 4  Next
Author Message
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #21 Posted: Sat Sep 01, 2018 2:10 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
zermelo wrote:
I'm quite sure that we humans cannot learn zero-style (though for a devil's advocate, has this actually been proven...?).

I think we could, and with fewer games than computers as we are better at learning/generalising from small data sets. But LZ has taken 9 million games to get to where she is today and humans tend to die before they can play that many games. I've sometimes wondered what would happen if 2 people were stranded on a desert island with a go board (and just knew the rules) for years and just played against each other, what would their go look like?

Here's an analysis of problem 87 I just opened the book at. The stated aim is "How can black make thickness for his stones?". I thought this shape could result from a strangely ignored outside approach to 3-4, but pattern searching shows it as a 6-4 joseki. There is one pro game that fits with the hoshi extension: Onoda Takuya 3p (black) vs Murakawa Daisuke 8p in 2017. Onoda played the squeeze starting with q18 which I correctly expected to be the book answer as it's a classic example of good move order to get maximum forcing moves. LZ doesn't like leaving high corner moves open so slightly preferred tenuki to add a move there.
Attachment:
MGS p87.PNG
MGS p87.PNG [ 702.64 KiB | Viewed 11596 times ]


If I turn the lower right corner into a 4-4 then it wants to continue locally, but with the book wrong answer of just p17 atari and then r14 sente block (omit s15 sente, interesting) and then p12 knight connection (initially preferred o12) instead of p14 solid in book. The book sequence would leave black at 32.8% on this board, whilst LZ's is 41.5%. 1% of that improvement comes from not playing s15, but it's mostly the more efficient knight move defence of the cutting point.
Attachment:
MGS p87 a.PNG
MGS p87 a.PNG [ 208.92 KiB | Viewed 11596 times ]


How does LZ like the book sequence? It's actually pretty close behind. With the s15 atari and solid p14 connect black is 37.3% which is > 32 so LZ agrees the book's right answer is better than the book's wrong answer. However, LZ would prefer to not s15 and then make the same efficient knight move defence for a 40.3%, just a little worse than p17 atari version. It's easy to see that although the q18 sacrifice allows black to seal the top side on the 2nd line, it gives white a stone at p17, reducing the liberties of the p16 stones and makes a cutting point at o16. Despite that weakness LZ still doesn't want to solid connect, but it's logical that if you are going to solid connect, it's more okay to play the q18 sacrifice because solid connect reduces o16 cut bad aji.
Attachment:
MGS p87 b.PNG
MGS p87 b.PNG [ 220.37 KiB | Viewed 11596 times ]


So my conclusion:
- the increased efficiency of defending the cut with knight move instead of solid connect is most important issue and book ignored it (though it did say "make thickness" so maybe that can be interpreted to mean don't play the best efficient move but play more solidly even if it's worse)/
- closing off the top side isn't as important as black getting a strong wall at p17 without o16 cut (in this position at least, and giving black f17 doesn't change that).
- think before s15 atari (it's a classic tseuji to make white s14 gote, but if white's not going to s14 soon do you need to play it now? it loses a ko threat and some yose compared to t15 if black gets to play first there)

P.S. Without extra side stone pro games are about half and half between the sacrifice squeeze and simple atari, e.g. http://ps.waltheri.net/database/game/70665/

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #22 Posted: Sat Sep 01, 2018 2:48 am 
Dies in gote

Posts: 46
Liked others: 109
Was liked: 34
Rank: Euro 1 dan
GD Posts: 7
Uberdude wrote:
zermelo wrote:
I'm quite sure that we humans cannot learn zero-style (though for a devil's advocate, has this actually been proven...?).

I think we could, and with fewer games than computers as we are better at learning/generalising from small data sets. But LZ has taken 9 million games to get to where she is today and humans tend to die before they can play that many games. I've sometimes wondered what would happen if 2 people were stranded on a desert island with a go board (and just knew the rules) for years and just played against each other, what would their go look like?


Right. I suppose I just meant that zero type learning would probably be much less effective than learning from better players in some way. But OTOH, maybe practising tsumego or endgame in zero style could be more effective (for improvement) than spending time to learn to imitate pro fuseki.


Last edited by zermelo on Sat Sep 01, 2018 3:07 am, edited 1 time in total.
Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #23 Posted: Sat Sep 01, 2018 3:02 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
We seem to have agreed that bots tenuki more than human pros, and seem to like sente (hence also forcing moves) more.

What else? It's only an impression, because running a bot to check a human problem is so tedious, but it seems to me that bots generally agree that the humans have good suji - they get the flow of stones right - but they often get the first move wrong. In one problem I just looked at, LZ accepted an 18-move non-obvious (to me) pro sequence once move 1 was played - but it didn't like Black's move 1. The human comment was that was necessary otherwise Black ended up being counterattacked, but the bot was unfazed by the counterattack. From other examples, too, I get the sense that LZ evaluates the safety of groups quite differently from humans (not in a different way, necessarily, but certainly with different confidence levels). I'm guessing that it sees resources in reserve that humans either miss or are unsure about. In other words, it is effectively reading deeper.

Does that sound right to you? If it is, on the one hand it seems a little depressing - we'll never match depth even with a string of new proverbs - but on the other hand it means we can probably keep taking the suji pills prescribed by the pro teachers. It's mainly their evaluations we have to wary of.

Musing further, I suggest the only sane way to have some sort of confidence in your own evaluations (in the absence of obvious blunders, of course) is not to say "what would a pro play" but simply to ask "how consistent is this with my previous play?" That's essentially means choosing a style. We all think we do that, but in practice how often do we amateurs diverge, veering off to chase the clever bird that shoots out of the undergrowth, taking us away from its nest and we end up falling in a bog?

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #24 Posted: Sat Sep 01, 2018 6:59 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
zermelo wrote:
Uberdude wrote:
zermelo wrote:
I'm quite sure that we humans cannot learn zero-style (though for a devil's advocate, has this actually been proven...?).

I think we could, and with fewer games than computers as we are better at learning/generalising from small data sets. But LZ has taken 9 million games to get to where she is today and humans tend to die before they can play that many games. I've sometimes wondered what would happen if 2 people were stranded on a desert island with a go board (and just knew the rules) for years and just played against each other, what would their go look like?


Right. I suppose I just meant that zero type learning would probably be much less effective than learning from better players in some way.


I am of the school that leans towards learning from better players by playing against them. Not that you can't learn from a book or lecture, but most learning is caught, not taught. :)

Quote:
But OTOH, maybe practising tsumego or endgame in zero style could be more effective (for improvement) than spending time to learn to imitate pro fuseki.


I have studied the endgame zero style and imitated pro fuseki. My study of tsumego has been a combination of both. IMX, imitation wins hands down for early learning. Up to where you have to start unlearning. In chess there are certainly geniuses like Capablanca, Morphy, and Fischer who did not rely much upon imitation but leaped ahead of everybody else.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #25 Posted: Sat Sep 01, 2018 7:39 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
John Fairbairn wrote:
We seem to have agreed that bots tenuki more than human pros, and seem to like sente (hence also forcing moves) more.


I think you are addressing Uberdude, but if I may. . . . :) I agree that bots tenuki more, and so perhaps will play sente as part of that, or in preparation for tenuki.

Quote:
What else? It's only an impression, because running a bot to check a human problem is so tedious, but it seems to me that bots generally agree that the humans have good suji - they get the flow of stones right - but they often get the first move wrong. In one problem I just looked at, LZ accepted an 18-move non-obvious (to me) pro sequence once move 1 was played - but it didn't like Black's move 1.


Interesting. :)

Quote:
The human comment was that was necessary otherwise Black ended up being counterattacked, but the bot was unfazed by the counterattack.


Right. That's my impression of bot style, as well. What? Me worry? ;)

Quote:
From other examples, too, I get the sense that LZ evaluates the safety of groups quite differently from humans (not in a different way, necessarily, but certainly with different confidence levels).


Yes. This is why, I think, the bots don't often bother to make traditional human bases for their stones.

Quote:
I'm guessing that it sees resources in reserve that humans either miss or are unsure about. In other words, it is effectively reading deeper.


Well, certainly the bots produce larger game trees than humans, both broader and deeper. (It is possible that Leela Zero reads too deeply, and not broadly enough. :scratch: ) But it is not the calculation of variations that has made Leela Zero strong and keeps it improving, it is its evaluation of positions and choice of candidate plays. Improving them is what enables it to see better.

Quote:
Does that sound right to you? If it is, on the one hand it seems a little depressing - we'll never match depth even with a string of new proverbs - but on the other hand it means we can probably keep taking the suji pills prescribed by the pro teachers. It's mainly their evaluations we have to wary of.


About the depth of reading of the bots. Their reading is global, and sometimes human local reading trumps it. A bot may judge a group to be safe while a human will see how to kill it or make a ko. This is also why, in the more nebulous cases where a bot will tenuki instead of making a local play to protect a group, I don't think it is that the bot has necessarily read the local situation more deeply than a human pro can. Usually I think it just has better judgement.

I agree that humans are not going to develop their calculation of variations much beyond what they are capable of at age 20 or so, but, just as Leela Zero and other bots continue to improve by improving their choice of candidate plays and their evaluation of positions, IOW, their judgement and intuition, humans can, too. :D

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #26 Posted: Sat Sep 01, 2018 11:09 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Uberdude wrote:
Here's an analysis of problem 87 I just opened the book at. The stated aim is "How can black make thickness for his stones?". I thought this shape could result from a strangely ignored outside approach to 3-4, but pattern searching shows it as a 6-4 joseki. There is one pro game that fits with the hoshi extension: Onoda Takuya 3p (black) vs Murakawa Daisuke 8p in 2017. Onoda played the squeeze starting with q18 which I correctly expected to be the book answer as it's a classic example of good move order to get maximum forcing moves. LZ doesn't like leaving high corner moves open so slightly preferred tenuki to add a move there.

Image

For one thing, I don't think that we have to strain after gnats in terms of Leela Zero's evaluation. All of these are possible. What about the keima to O-18? It closes off this entrance to the top side with sente, but without sacrificing a stone. :) OTOH, assuming that White plays Q-18, it gives White possibilities that the sacrifice does not, such as N-17.

Quote:
If I turn the lower right corner into a 4-4 then it wants to continue locally, but with the book wrong answer of just p17 atari and then r14 sente block (omit s15 sente, interesting) and then p12 knight connection (initially preferred o12) instead of p14 solid in book. The book sequence would leave black at 32.8% on this board, whilst LZ's is 41.5%. 1% of that improvement comes from not playing s15, but it's mostly the more efficient knight move defence of the cutting point.

Image

This keima connection is very nice. I would have chosen the book sequence, with the one stone sacrifice and the solid connection, but Leela Zero judges that to be bad. IOW, the value of this keima is worth more than sealing off the open skirt. How does the extension affect that decision?

Quote:
How does LZ like the book sequence? It's actually pretty close behind. With the s15 atari and solid p14 connect black is 37.3% which is > 32 so LZ agrees the book's right answer is better than the book's wrong answer. However, LZ would prefer to not s15 and then make the same efficient knight move defence for a 40.3%, just a little worse than p17 atari version. It's easy to see that although the q18 sacrifice allows black to seal the top side on the 2nd line, it gives white a stone at p17, reducing the liberties of the p16 stones and makes a cutting point at o16. Despite that weakness LZ still doesn't want to solid connect, but it's logical that if you are going to solid connect, it's more okay to play the q18 sacrifice because solid connect reduces o16 cut bad aji.

Image

My inclination would have been to make the solid connection because of the sacrifice, not the other way around. ;)

Quote:
So my conclusion:
- the increased efficiency of defending the cut with knight move instead of solid connect is most important issue and book ignored it (though it did say "make thickness" so maybe that can be interpreted to mean don't play the best efficient move but play more solidly even if it's worse)/


When I was 3 kyu or so I read a book on shape and it really opened my eyes. I became a shape player until I reached 4 dan. (I had to give up my attachment to good shape in order to do so. ;)) My personal definition of shape is local efficiency. OC, the keima connection is quite efficient, so I'm surprised they were tacit about it.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #27 Posted: Sat Sep 01, 2018 11:55 am 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
LZ doesn't like leaving high corner moves open so slightly preferred tenuki to add a move there.


Without intending to contradict that, I'd like to toss in question that could lead to a different interpretation.

In all the copious literature on thickness, I've seen next to nothing that talks about the timing of thickness. Thickness early on lets you make grand plans and maximise extensions (emptier board) but you are showing your hand very clearly and your opponent can dance around it. Also, thickness needs connections but early in the game you can't often tell whether a keima, tiger's mouth or solid connection is best, and this kind of thickness runs the risk of being overconcentrated. The converse more or less applies later in the game, but there is perhaps also greater risk of disruption.

So dies it matter - early or late? I ask this because among the many vague impressions I'm forming one is that bots don't like making thickness early on. Indeed they like giving it, as in the now notorious 3-3 invasion!

The Murakawa-Onoda game cited above was a case of very early thickness, and my impression was that Black just couldn't make that thickness do anything useful (he resigned early, too). In that light, the bot's choice of tenuki-ing to the shimari makes some sort of sense even to me. Do you think the bot's choice was not a hierarchy of one plan - make the shimari - but a hierarchy of two - 1. avoid early thickness, 2. make the shimari?

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #28 Posted: Sat Sep 01, 2018 12:37 pm 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Going back to zermelo's points about how much we can trust our human experts, best way to learn etc. With super-strong AI it's now easy for weaker players to discover mistakes in what they used to take on trust from stronger ones, and some of the aura of strength will fade. But I'd still be happy to play as 'badly' as Rob van Zeijst 7d or some Japanese 9p from yesteryear who LZ says makes big mistakes nearly every move in the opening. In some specific situations where I have newer knowledge informed by more recent pro or AI thinking I can play better than them, but overall I'm still far weaker because the most important thing for one's strength is not how good your best moves are but how bad are your worst (and reading power). And I still think the best way to pull up the level of your worst moves is tsumego, so if I just wanted to get stronger I'd do that, but I play more for interest and fun.

I'm also reminded of a correspondence game I played on OGS some years ago, in which we played an avalanche variation and then another complicated joseki in the adjacent corner identically to a pro game (I'd looked it up and presume my opponent had too). Trouble was I was playing the side of the pro who lost, so I couldn't keep imitating for ever! So I studied the pro game with the aim of finding his mistakes so I could improve on his result (which might sound presumptuous, but I had his pro opponent to help). As it happens I managed this (with some help from my opponent) and now we've got LZ I think I'll analyze my game compared to the pro's in my journal.

Edit for posts made whilst I was drafting this:

John: Yes, I agree there is a tendency of bots to not be so impressed by early thickness, if we can call it that (as with the 3-3 invasion it's a wall but it's not yet alive). And they do seem very good at finding the right depth of reduction later. But I think with the shimari it is mostly about not liking high initial corner moves as I've seen this kind of wanting to come back to close it early in many positions, and it didn't mind making the thickness if the other corner was a 4-4 instead of "urgent to correct the mistake" 6-4.

Bill: If black plays the knight move to 2nd line LZ white would s16 capture to avoid r14 sente (p17 push to create cutting point beforehand possible) but definitely not q18.

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #29 Posted: Sat Sep 01, 2018 1:13 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Uberdude wrote:
I still think the best way to pull up the level of your worst moves is tsumego, so if I just wanted to get stronger I'd do that, but I play more for interest and fun.


At this point I think that for DDKs the best way to pull up the level of their worst moves is to study damezumari at the end of the game and the lurking semeai. Big swings happen there. ;) Otherwise, I would use a strong bot and see which moves lose the most percentage points.


Quote:
Bill: If black plays the knight move to 2nd line LZ white would s16 capture to avoid r14 sente (p17 push to create cutting point beforehand possible) but definitely not q18.


Good point. Thanks. :) Does that hold true in response to O-17 as well? (O-17 and O-18 were not on my radar. ;))

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #30 Posted: Sat Sep 01, 2018 1:39 pm 
Dies in gote

Posts: 46
Liked others: 109
Was liked: 34
Rank: Euro 1 dan
GD Posts: 7
Uberdude wrote:
Going back to zermelo's points about how much we can trust our human experts, best way to learn etc. With super-strong AI it's now easy for weaker players to discover mistakes in what they used to take on trust from stronger ones, and some of the aura of strength will fade. But I'd still be happy to play as 'badly' as Rob van Zeijst 7d or some Japanese 9p from yesteryear who LZ says makes big mistakes nearly every move in the opening.


Yes, I agree with this completely, and that is why I suggested that we can still use the better players sequences as positive examples to teach our personal 'policy networks', maybe even when we don't understand the reasoning behind the moves. The bots have not changed anything in this regard.

What I think the bots have shown is that even the strongest players have been too opinionated about 'bad' positions even when they in fact don't understand them very well. Like the book in question that is full of 'failures' that are in fact perfectly valid sequences. Then the weak players like me have spent too much time trying to learn to replicate these judgements. I think we can learn something useful from this.

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #31 Posted: Sat Sep 01, 2018 4:47 pm 
Oza

Posts: 3647
Liked others: 20
Was liked: 4626
Quote:
What I think the bots have shown is that even the strongest players have been too opinionated about 'bad' positions even when they in fact don't understand them very well.


I don't think the bots have shown this as something new. One of the main characteristics of the series of books of commented game I have done (Kamakura, etc) is that the commentaries have been compiled on the basis of many pro sources, and I have shown many times that pros can disagree vehemently about all sorts of moves.

I have also demonstrated very many cases (usually thanks to Go Seigen) in Gateway To All Marvels where answers to tsumego problems by 9-dans are flawed.

There have been cases where 9-dans have made spectacular booboos in endgame problems (one by Kano Yoshinori spotted by Matthew Macfadyen comes to mind. I'm sure people can point to other examples, and of course we mustn't overlook the innumerable cases where a pro has made what he believed to be a good move only for it to turn out badly, and the likewise common cases where pro commentators claim to have found mistakes by even Go Seigen and Dosaku.

(But in the case of the problems here and similar books, I'm not convinced that the bots are necessarily showing even the strongest players are wrong, for the simple reason that there's a good chance that the books were in fact written by amateur ghost writers. Plus, of course, the book here is not about making the best move but about making good shape, which may be something rather different.)

Anyway, pros getting it wrong and this being pointed out is nothing new.

Yet we all feel the bots are telling us something new. Why, and what is it? I'm not at all. Maybe it's something to do with the frequency of apparent pro mistakes. Maybe they are not mistakes at all but a matter of style. If we allowed humans to play in such a way that they could somehow not make blunders and could play each move with a consistency that escapes them because of fatigue, carelessness or whatever - e.g. they could take back moves - would they sometimes beat bots (there have been enough close games with bots to suggest a small boost in the human style of play could earn victories).

Clearly, no-one knows the answers yet, but I think it's agreed that we now know humans must learn to consider more candidate moves. The other area we need to consider is evaluation. I would like to suggest a novel approach to that which I found by accident.

I too have tried posing problems from books to LZ, but the way I did it was to input the positions in a haphazard order and not in the order of play. What I noticed was that if you put, say, a move on the second line that had no proper relationship to the other stones, the winrate graph went berserk, but when you later add another stone which makes that stone relate sensibly to other stones, the winrate graph would flatten out again.

I then experimented a bit, and made a point of trying to randomise the input. What I further discovered is that you can effectively ignore the winrate. It is inherently flat anyway, because the position once completely input is close to level (being from a pro game). But if you instead look at the variation lines (which hitherto I have tended to ignore, and I infer others do, too), you can see how LZ is "thinking."

To give a specific example, we put Black and White stones down alternately with pondering "on" as you do this, and let White make the most sensible move suggested by LZ (but initially I kept White stones away from the Black ones). For Black we put down S3 then Q8 as the first moves in the lower right quadrant (there may be Black moves elsewhere, though). We essentially ask LZ how to "repair" this awful shape. It suggests (not necessarily in the very next moves being considered but in the variation lines shown) that N3 is a move it likes for Black. Now add another Black stone at S7 and again ask LZ how to repair this daft shape. This time it suggests M3 (or M4). We can infer several things from this. It appears to "think," for example, that the stone at S7, daft as it is, is adding enough strength to justify a bigger extension. It's not really saying that, of course, but it's demonstrating something in a way a human like me can talk about, both in terms of what I already know (that you extend from strength according to how strong the position is) but also in terms of what I don't know - or just don't think about - namely that the weird position of S7 and S2, which I would regard as so daft and useless as to be not worth thinking about, actually adds enough strength to justify a wider extension.

If you then add a White stone at N3 (stopping any extension), LZ flips rapidly between L3 (pincer) and P3 (tsume) for Black in the variation lines, creating a feeling that is "talking" to me, saying "I can't make my mind up about these two moves."

If you then add another daft Black stone at K5 and look at the later variations where Black gets to move first in this area, we see that it now clearly prefers tsume at P3. LZ apparently does not want to try to enclose the White stone and build thickness by bullying from the outside (which is what I would have done).

I don't have the time to examine these extreme positions systematically, but I can say from the little I've done that I do feel LZ is now talking to me in terms I can just about follow. Maybe others would like to experiment with this technique?


This post by John Fairbairn was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #32 Posted: Sat Sep 01, 2018 6:53 pm 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
John Fairbairn wrote:
But if you instead look at the variation lines (which hitherto I have tended to ignore, and I infer others do, too), you can see how LZ is "thinking."
{snip}

I don't have the time to examine these extreme positions systematically, but I can say from the little I've done that I do feel LZ is now talking to me in terms I can just about follow.


Wonderful and fascinating, John1 :clap: :bow: :clap:

I am reminded about how much we have discovered about how the brain works by seeing how it acts after being damaged, by disease, strokes, or accidents. More power to you. :D

Edit: Or how much we learn about human vision from optical illusions. ;) We don't have to assume that Leela Zero is right about the plays it comes up with in bizarre situations. After all, it has not been trained on them. But we can learn something about how LZ works. :) Brilliant, John! :clap: :salute: :bow:

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #33 Posted: Sun Sep 02, 2018 3:15 am 
Lives in gote

Posts: 426
Liked others: 186
Was liked: 191
I decided to put a few of the problems from 'Attacking and Defending Moyos' (same authors) into LZ. Leela Zero does find other moves it prefers over the book answers and there is one terrible move in the solution I have found so far, but sometimes I get the feeling this is partly because in the positions chosen it evaluates black as being significantly ahead (or behind) at the start of the position. This affects everything for a bot, maybe more so than for a human. The book answer should be pretty easy for players who have studied problems like this, but white's response surprised me a bit. And I would not have guessed where LZ wanted to play as black in this position, either.

I should say that while I enjoy this exercise, I respect the authors. They have done a great service for the English-speaking go community over many years, and so I do not intend to post more positions. Books like this teach a way of thinking, and practical ideas for amateurs, and do not presume to reveal the absolute truth. LZ also teaches a way of thinking that we can't completely understand, but it is also likely far from the truth.

Click Here To Show Diagram Code
[go]$$B Problem 4. Black to play and expand moyo.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . O . . . . O . . . . . |
$$ | . . O , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . O . . . . . |
$$ | . . . . . . . . . . . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . X . . . . . . |
$$ | . . . O . X . . . , . . . . X X X . . |
$$ | . . . . . . . . . . . X . O . O O . . |
$$ | . . . . . . . . . . . . . O . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #34 Posted: Sun Sep 02, 2018 3:47 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
My policy network for this position (I expect a is the book answer, but I'd like to get b in sente first if I can). h is for giggles.
Click Here To Show Diagram Code
[go]$$Bc Problem 4. Black to play and expand moyo.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . e . . . . O . . . . O b . . . . |
$$ | . . O d c . . . h , . . . . . X . . . |
$$ | . . . . . . . . . . . . . W . . . . . |
$$ | . . . . . . . . . . . . . i . X . . . |
$$ | . . . . . . . . . . . . . a . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . f . . . . . . . . . . . . . . . |
$$ | . . . O . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . X . . . . . . |
$$ | . . . O . X . . . , . . . . X X X . . |
$$ | . . . g . . . . . . . X . O . O O . . |
$$ | . . . . . . . . . . . . . O . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

I just checked Elf v1, and it likes b and f most. f is only available because white made a one-space jump there instead of 2 space, which Elf would prefer a bit. With this kind of Takemiya moyo building you sometimes see black making the 2-space high approach and then 2-space jump in that corner. But most interesting thing for me was what Elf preferred for white's last move instead of the marked jump: p12 reduction. That kind of move is rarely on my radar but now that I think about it I recall Matthew Macfadyen 6d suggesting such ideas in similar positions.

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #35 Posted: Sun Sep 02, 2018 4:15 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Uberdude wrote:
My policy network for this position (I expect a is the book answer, but I'd like to get b in sente first if I can). h is for giggles.
Click Here To Show Diagram Code
[go]$$Bc Problem 4. Black to play and expand moyo.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . e . . . . O . . . . O b . . . . |
$$ | . . O d c . . . h , . . . . . X . . . |
$$ | . . . . . . . . . . . . . W . . . . . |
$$ | . . . . . . . . . . . . . i . X . . . |
$$ | . . . . . . . . . . . . . a . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , . . . . . , . . . . . X . . . |
$$ | . . . f . . . . . . . . . . . . . . . |
$$ | . . . O . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . X . . . . . . |
$$ | . . . O . X . . . , . . . . X X X . . |
$$ | . . . g . . . . . . . X . O . O O . . |
$$ | . . . . . . . . . . . . . O . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]

I just checked Elf v1, and it likes b and f most. f is only available because white made a one-space jump there instead of 2 space, which Elf would prefer a bit. With this kind of Takemiya moyo building you sometimes see black making the 2-space high approach and then 2-space jump in that corner. But most interesting thing for me was what Elf preferred for white's last move instead of the marked jump: p12 reduction. That kind of move is rarely on my radar but now that I think about it I recall Matthew Macfadyen 6d suggesting such ideas in similar positions.


Click Here To Show Diagram Code
[go]$$Bc Problem 4. Black to play and expand moyo.
$$ ---------------------------------------
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . e . . . . O . . . . O b . . . . |
$$ | . . O d c . . . h , . . . . . X . . . |
$$ | . . . . . . . . . . . . . W . . . . . |
$$ | . . . . . . . . . . . . . i . X . . . |
$$ | . . . . . . . . . . . . . a . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . , 1 . . . . , . . . . . X . . . |
$$ | . . . f . . . . . . . . . . . . . . . |
$$ | . . . O . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . X . . . . . . |
$$ | . . . O . X . . . , . . . . X X X . . |
$$ | . . . g . . . . . . . X . O . O O . . |
$$ | . . . . . . . . . . . . . O . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ ---------------------------------------[/go]


As an expanding moyo move, I expect that "a" is the book answer. But I was glad to hear that Elf likes "f", because in a real game I have the suspicion that I would play :b1:. Viewed from the standpoint of mutual damage, I think it does more damage.

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #36 Posted: Sun Sep 02, 2018 5:00 am 
Dies in gote

Posts: 46
Liked others: 109
Was liked: 34
Rank: Euro 1 dan
GD Posts: 7
John Fairbairn wrote:
Quote:
What I think the bots have shown is that even the strongest players have been too opinionated about 'bad' positions even when they in fact don't understand them very well.


I don't think the bots have shown this as something new. One of the main characteristics of the series of books of commented game I have done (Kamakura, etc) is that the commentaries have been compiled on the basis of many pro sources, and I have shown many times that pros can disagree vehemently about all sorts of moves.

I have also demonstrated very many cases (usually thanks to Go Seigen) in Gateway To All Marvels where answers to tsumego problems by 9-dans are flawed.


Fair enough, John. Maybe this is not new for everyone but at least I have realised something new. Of course I’ve known too that pros have different opinions on joseki and whole board judgement etc. Maybe the new learning is just a quantitative difference. I now know better how much I should trust it when a strong player says something is ‘bad’.

Still I think this ties into what other stronger players have said. I recall Ke Jie said something like that he’s going to stop trusting any whole board intuition and only trust reading and calculation from now on, after the Alphago experience. And if I understood Fan Hui’s point in his Alphago lectures, where he likes to say that “You can play anything! Anything goes!”, he means exactly that many players have been too judgemental were they have no real understanding, and that Alphago taught something about this.

Regarding ‘making good shape’ and similar books, I can believe that I did not necessarily need a bot to make the same conclusion. Maybe if I had shown the positions to 10 pros that would give their honest opinions, they would have told that there are other ways to play, and the ‘failures’ are often quite ok. It is just not something I can easily do, and then people often don’t like to criticise there colleagues.

I want to add that I’m sure that the ‘Making good shape’ book has done much more good than harm to my go skills. I have just to some extent tried to use that kind of books incorrectly. In a life and death book where you can solve 70% of the problems, you can pretty much trust that you will be stronger when you can solve the last 30% too. I assumed that if I want to be strong I need to be able to solve all problems from ‘Making good shape’ too. Now I see that trying to get there was pointless.

I’ll reiterate my earlier conclusion: If a much stronger player says that something is good, you can trust that it is good enough for your games. If they say something is bad, don’t trust them, unless you can verify it by reading or calculation.

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #37 Posted: Sun Sep 02, 2018 6:21 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
So this thread doesn't come across as a van Zeijst / Bozulich bash, I took a look at a problem from "Shuko: The Only Move" vol 2. These are a series of whole board problems from his study group in which he suggests better moves than the young [relative to him!] pros played in their games, so we can be pretty sure it's really the thoughts of the great Fujisawa Shuko and not a ghostwriter.
Click Here To Show Diagram Code
[go]$$Bcm19 Problem 39: Takao Shinji 7p (black) vs Morita Michihiro 9p
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O O . O O O . O . . . . . . . . |
$$ | . . . , X . . X X X . . . . . , X . . |
$$ | . . X . . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . O . 1 . . . . . . . . . . . |
$$ | . . . , . . 2 3 . , . . . . . , . . . |
$$ | . . . X b 7 . 6 . . . . . . . . . . . |
$$ | . . . c a . 4 . . . . . . . . . . . . |
$$ | . . d . . . . . . . . . . . . . . . . |
$$ | . . . 5 . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . , . . . . . O . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]


Black has just peeped with 25. The question is should you just connect or resist? In the game Morita simply connected, whilst Shuko says counter-peeping at a is the only move. The point is if black connects at b that's a good exchange to make black heavy and set up d for later, whereas if you connect directly then later if you peep black will not connect but treat the peeping stone lightly (it's already done its job of reducing white's eyeshape) and just defend at c. This is a pretty common technique (and I managed to get this problem right, others are harder). Elf v1 basically agrees with this logic: connect is the instinct but counter-peep quickly is found and has higher win% but I didn't run it long enough on my weak computer to overtake in playouts. If black does connect to the counterpeep white gains 7% compared to directly connecting. Incidentally Elf has some quite different ideas to the pros about the flow of the fight following the h11 cap, basically it doesn't like the cap (prefers tenuki e.g. lower right) and then doesn't want to run (Shuko mentions e9 and f9 as possibilities as well as the kosumi, one of these happened in his Kisei title match) but just settle inside with c13.

In the actual game Morita simply connected. He did consider the counter-peep but was worried about a black resistance shown below. Shuko says it's complicated but playable for white as e breaks out if black attempts the fence of d. Elf agrees with Shuko this result is good for white, but sees quite a few better moves in the sequence, for example it sees white e5 extend as a huge 50% mistake because it's not sente and black f10 connect in reply would put white in big trouble (playing around with the variations of white the d5 push and cut it seems legit to me). Or e10 at e7 for sacrifice 10% diff.

Click Here To Show Diagram Code
[go]$$Wcm26 Morita's worry
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O O . O O O . O . . . . . . . . |
$$ | . . . , X . . X X X . . . . . , X . . |
$$ | . . X . . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . O . X . . . . . . . . . . . |
$$ | . . . 5 4 c O X . , . . . . . , . . . |
$$ | . . . X 3 X . O . . . . . . . . . . . |
$$ | . . . 2 1 6 O . . . . . . . . . . . . |
$$ | . . . 0 7 8 . . . . . . . . . . . . . |
$$ | . . . X 9 e . . . . . . . . . . . . . |
$$ | . . . b a . . . . . . . . . . . . . . |
$$ | . . . X . d . . . , . . . . . O . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]


Takao said he would have cut in response to the counter-peep. Shuko says that's no good as white is good with the squeeze, Elf strongly agrees (30% mistake) and agrees with the one-way-street given.

Click Here To Show Diagram Code
[go]$$Wm26 Takao's plan
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O O . O O O . O . . . . . . . . |
$$ | . . . , X . . X X X . . . . . , X . . |
$$ | . . X . . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . O . X . . . . . . . . . . . |
$$ | . . . , . 9 O X . , . . . . . , . . . |
$$ | . . . X 3 X 2 O 5 . . . . . . . . . . |
$$ | . . . . 1 6 0 4 . . . . . . . . . . . |
$$ | . . c . a 7 8 . . . . . . . . . . . . |
$$ | . . . X . . b . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . X . . . . . , . . . . . O . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]


So conclusion so far:
- Elf agrees with Shuko counter peep is better (easily so if black obeys)
- Elf agrees with Shuko that the end result of Morita's worry of black resistance is fine for white (but some big mistakes from both in the sequence Shuko didn't highlight)
- Elf agrees with Shuko that Takao's planned counter was bad

Pretty good marks so far. But Elf sees a better black counter that isn't mentioned in the book. The push and then connect. My explanation is that black also makes white heavy and by cutting on the outside delays that c7 weakness by starting a new fight so white is too busy (though the a-c tesuji is incoming and there's lots of violence).

Click Here To Show Diagram Code
[go]$$Bcm27 Elf black best answer to counterpeep
$$ +---------------------------------------+
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O O . O O O . O . . . . . . . . |
$$ | . . . , X . . X X X . . . . . , X . . |
$$ | . . X . . X . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . O . O . X . . . . . . . . . . . |
$$ | . . . , . . O X . , . . . . . , . . . |
$$ | . . . X 3 X 4 O . . . . . . . . . . . |
$$ | . . . . O 1 O . . . . . . . . . . . . |
$$ | . . b a . 2 5 7 . 9 . . . . . . . . . |
$$ | . . c X . 6 . . . . . . . . . . . . . |
$$ | . . . . . . 8 . . . . . . . . . . . . |
$$ | . . . X . . . . . , . . . . . O . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ | . . . . . . . . . . . . . . . . . . . |
$$ +---------------------------------------+[/go]


A quick summary of problem 40. Elf agrees with Shuko that young Kenmochi's play was bad: Shuko: "black's profit is very big ... discouraging for white to continue ... a little painful and difficult; Elf: "first move -40%, ends with white 4%". Elf thinks Shuko's proposed move is ok (some diagreement over following sequence), but another move a few % better.

Problem 42 is the third I look at and first where Elf strongly disagrees with Shuko's sequences and conclusions. Shuko says Takao's play as white was very heavy and his opponent just played obvious moves for a better result. Elf says Takao started at 35%, his first move was indeed bad (down to 22) but his opponent made a bigger mistake (unremarked by Shuko, I would have played Elf's move!) and Takao's bad result ended up good at 70%. On the other hand Shuko's only move was -17% and his mainline result of living in sente was indeed better (white 52% with where he spent his sente, 66% with Elf's) but contained indulgent reading of soft play by opponent so was really 45% says Elf. He did offer an alternative reply to his 1st move, Elf says much better, and ends with a result he sees as ok "easy for white to make sabaki" which Elf hates (white 4%, and this is with more komi than the games presumably had). The AIs are rather territorial so there could be a style difference in that humans can win with a light sabaki result giving the opp territory, whereas bot likes the cash, but a Shuko preferring a 4% result over Takao's 70% is quite a discrepancy!


This post by Uberdude was liked by: Bill Spight
Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' problems.
Post #38 Posted: Sun Sep 02, 2018 9:05 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Uberdude wrote:
So this thread doesn't come across as a van Zeijst / Bozulich bash, . . .


I expect that the material in their shape book is standard fare, with little in the way of original composition. They may have altered the contexts of problems so as not to be accused of copying. So any criticism is not directed at them personally, but at generally accepted go "knowledge". For instance, the underneath connection appears in Sakata's Tesuji and Antisuji of Go, with a slightly different context. It is presented as a connection problem, so the hane is a failure, but it also judges the hane harshly, saying that it causes great damage to White's position. Leela Zero offers a valuable correction. :)

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' and other boo
Post #39 Posted: Sun Sep 02, 2018 9:46 am 
Judan

Posts: 6725
Location: Cambridge, UK
Liked others: 436
Was liked: 3719
Rank: UK 4 dan
KGS: Uberdude 4d
OGS: Uberdude 7d
Another Shuko problem makes for an interesting "find the pro's mistakes" exercise. The problem is at move 60, but in the intro to the position Shuko picks out one black move for harsh criticism. Elf agrees Shuko's move is better (but not best, that's one puzzlingly similar to the bad move), but picks out 1 white move as being terrible (-26%), can you find it? Playing through the game it did seem odd to me and I expected the Elf move. Elf also highlights another large (10%) white mistake but that one's rather more subtle but made sense to me once I'd played around with some variations.

So whilst Elf agrees Takao's 60 was bad and likes Shuko's suggestion (though expected variation different), it's saying he's focusing on the wrong part of the game as black was winning with 99.6% just a few moves earlier.



(Leela Zero #157 agrees the terrible move was bad (just 7% in milder LZ land) and the other white mistake, but thinks Takao's 60 was fine, as is Shuko's alternative.)

Top
 Profile  
 
Offline
 Post subject: Re: Leela Zero analysis of 'Making good shape' and other boo
Post #40 Posted: Sun Sep 02, 2018 10:50 am 
Honinbo

Posts: 10905
Liked others: 3651
Was liked: 3374
Uberdude wrote:
Another Shuko problem makes for an interesting "find the pro's mistakes" exercise.


Just think. In ten years we can have "find the bot's mistakes" exercises. :lol:

_________________
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins

Visualize whirled peas.

Everything with love. Stay safe.

Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 67 posts ]  Go to page Previous  1, 2, 3, 4  Next

All times are UTC - 8 hours [ DST ]


Who is online

Users browsing this forum: No registered users and 1 guest


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group